Sanjeev Ranjan

Problem

PMOs and BAs spend hours after every meeting manually turning raw notes into requirements, user stories, acceptance criteria, and risks.

Solution

A 4-stage AI pipeline (requirements → stories → acceptance criteria → risks) that generates a structured, ID-linked requirements package from raw meeting notes, powered by Google Gemini.

Benefit

Turns a multi-hour documentation task into a reviewable first draft in seconds, complete with a live traceability matrix and flagged ambiguities that still need a human decision.

Problem

Same documentation bottleneck — but also a question of whether the same solution can be built reliably on more than one AI platform.

Solution

The identical 4-stage pipeline and traceability approach, built as a self-contained Claude Artifact that calls the Claude API directly from the browser — no separate backend.

Benefit

Demonstrates the same working solution shipped on two different AI platforms and architectures, from a single lightweight in-browser build.

Problem

Translating business requirements directly into working software usually needs a traditional dev handoff, slowing delivery and diluting PM oversight along the way.

Solution

Built a full-stack expense tracking app end-to-end using Claude Code as an AI development partner, applying a "big prompt" methodology — describing complete feature requirements upfront rather than incremental prompting. Next.js, TypeScript, and Tailwind CSS, with expense data persisted client-side via localStorage. Explored three parallel implementation approaches (simple export, advanced multi-format export, cloud-integrated export) using Git branching to compare options before selecting a direction — mirroring real portfolio evaluation and delivery decisions.

Benefit

Shows the ability to direct AI-driven delivery end-to-end — requirements, build, quality review, and version control — while retaining PM-level oversight (reviewing every file change, approving every command) throughout, not just prompt engineering in isolation.

Problem

Documentation debt — inconsistent, manually written, quickly-outdated docs — and the common failure of writing only for developers while end users are left without guidance.

Solution

Designed a reusable Claude Code slash command (/generate-docs) using a structured prompt-engineering framework (Task instructions, Arguments, Reusable process steps, Guided examples, Explicit outputs, Template naming, Error handling, Documentation). Given only a feature name, it analyzes the relevant code, classifies the feature as frontend/backend/full-stack, and generates two distinct, purpose-built documentation files: (1) a technical specification for developers — architecture, API details, and implementation notes — and (2) a user guide for non-technical readers — plain-language, step-by-step instructions with real captured screenshots. Both follow the project's existing documentation patterns, are cross-linked to each other, and are cross-referenced against related existing documentation.

Benefit

Demonstrates applying PMO governance instincts (standardization, dual-audience communication, traceability, built-in quality checks) directly to a technical AI workflow, rather than one-off AI usage.

Problem

Serial AI development is slow when multiple independent features are ready to build at once — and parallel work risks silently hiding the coordination points that still need human judgment.

Solution

Designed and ran a parallel AI development workflow combining Git worktrees with Claude Code subagents — spinning up multiple isolated working directories, each on its own branch, and assigning an independent AI subagent to build a full feature in each simultaneously (no shared state, no interference). Used this to build two features in parallel and surfaced a genuine shared-dependency conflict during integration (both features independently modified the app's navigation component in incompatible ways), requiring a manual reconciliation decision.

Benefit

Demonstrates orchestration of multi-agent technical delivery and the PMO skill of identifying and resolving shared-dependency conflicts between concurrent workstreams — a direct technical analogue to managing parallel project tracks.

Problem

Project status updates arrive scattered across emails, chats, and notes, making it hard to keep an accurate, current RAID log and executive status view.

Solution

Converts scattered project updates into one consolidated executive status report with RAG status and a structured RAID log.

Benefit

Replaces a manual weekly compilation task with an automated, always-current status view for leadership.

Problem

Key vendor contract terms — renewal dates, obligations, risk clauses — are easy to lose track of across dozens of SOWs and contracts.

Solution

Extracts key terms, renewal dates, and risk flags directly from vendor contracts and SOWs.

Benefit

Reduces missed renewals and contract risk by surfacing critical terms automatically instead of relying on manual contract review.